Point-based and model-basedgeolocation analysis of airborne laserscanning data

Umut Gunes SefercikGurcan BuyuksalihKarsten JacobsenMehmet Alkan

Umut Gunes Sefercik, Gurcan Buyuksalih, Karsten Jacobsen, Mehmet Alkan, “Point-based and model-basedgeolocation analysis of airborne laser scanning data,” Opt. Eng. 56(1), 013101 (2016),doi: 10.1117/1.OE.56.1.013101.

Point-based and model-based geolocation analysis ofairborne laser scanning data

Umut Gunes Sefercik,a Gurcan Buyuksalih,b Karsten Jacobsen,c and Mehmet Alkand,*aBulent Ecevit University, Department of Geomatics Engineering, Zonguldak 67100, TurkeybIMP Bimtas, Tophanelioglu Street, Istanbul 34430, TurkeycLeibniz University Hannover, Institute of Photogrammetry and Geoinformation, Nienburger Street 1, Hannover D-30167, GermanydYildiz Technical University, Department of Geomatics Engineering, Istanbul 34220, Turkey

Abstract. Airborne laser scanning (ALS) is one of the most effective remote sensing technologies providingprecise three-dimensional (3-D) dense point clouds. A large-size ALS digital surface model (DSM) coveringthe whole Istanbul province was analyzed by point-based and model-based comprehensive statisticalapproaches. Point-based analysis was performed using checkpoints on flat areas. Model-based approacheswere implemented in two steps as strip to strip comparing overlapping ALS DSMs individually in three subareasand comparing the merged ALS DSMs with terrestrial laser scanning (TLS) DSMs in four other subareas. In themodel-based approach, the standard deviation of height and normalized median absolute deviation were usedas the accuracy indicators combined with the dependency of terrain inclination. The results demonstrate thatterrain roughness has a strong impact on the vertical accuracy of ALS DSMs. From the relative horizontal shiftsdetermined and partially improved by merging the overlapping strips and comparison of the ALS, and the TLS,data were found not to be negligible. The analysis of ALS DSM in relation to TLS DSM allowed us to determinethe characteristics of the DSM in detail. © 2016 Society of Photo-Optical Instrumentation Engineers (SPIE) [DOI: 10.1117/1.OE.56.1.013101]

Keywords: airborne laser scanning; terrestrial laser scanning; height; digital surface model; accuracy.

Paper 160944 received Jun. 16, 2016; accepted for publication Oct. 31, 2016; published online Dec. 1, 2016.

1 IntroductionIn the last few decades, advanced airborne and space-borneremote sensing technologies have become indispensable,particularly for land-related professions requiring rapid,periodical, and high-accuracy three-dimensional (3-D) geo-information. Airborne systems offer higher resolution thanspace-borne methods since they can perform measurements atlower altitudes. Airborne laser scanning (ALS), in particular,can identify high-resolution point clouds using 3-D groundcoordinates. From the derived points, a digital surface model(DSM), which is the 3-D presentation of the entire vegetationand man-made objects in the observation area, and a digitalterrain model (DTM), which is related to the bare ground andinterpolates the earth surface and locations of buildings,are generated. DSM and DTM can be identified also withthe general inclusive term digital elevation model (DEM).With the advantage of providing 3-D and precise data, ALShas been adopted as one of the primary techniques formapping local to national scale coverage, a topic historicallydominated by photogrammetry1,2 and utilized in severalDEM-based studies.3–12

Considering the large utilization of ALS DEMs, their geo-location accuracy remains a significant question. This studywas designed to answer to this question by producing veryhigh resolution (VHR) ALS DSMs in several study areas,which have varied terrain formations, and validating themthrough visual and statistical approaches and geometric testsregarding the function of terrain slope. In ALS, the accuratedetection of geolocation depends on several parameters such

as flying altitude above ground, footprint size, and terrainroughness. In the literature, the quality of the ALS DSM ismostly assessed by a checkpoint-based comparison.13–18

However, checkpoints have special location attributes, whichusually do not represent the whole height model. For a com-plete and most reliable geolocation accuracy evaluationincluding all of the pixels of ALS raster DSMs, the check-point-based analysis was combined with model-to-modelapproaches that allow a 3-D-comparison of the ALS DSMsstrip to strip internally and the terrestrial laser scanning(TLS) DSMs.

This paper is organized as follows: Section 2 presentsinformation about the study areas in Istanbul and the materi-als used; it is followed by the methodology section, whichsummarizes the ALS DSM generation and accuracy valida-tion steps. The experimental results are presented and dis-cussed according to the mapping requirements in Sec. 4.Finally, Sec. 5 presents the conclusion of the study.

2 Study Areas and MaterialsThe greater municipality of Istanbul implemented a projectfrom 2012 to 2013 to collect ALS data over a 5400-km2 areato generate a 3-D model of the city based on a DSM andDTM. For a model-based quality analysis, seven study areaswith varying terrain characteristics were selected mainlyfrom the European side of Istanbul (Fig. 1). In test areas1 to 3, the overlap areas of the neighboring flight lineswere compared and in areas 4 to 7, the TLS data wereused as references. ALS was performed using a helicopterequipped with a Riegl laser scanner Q680i, and the TLS

*Address all correspondence to: Mehmet Alkan, E-mail: alkan@yildiz.edu.tr 0091-3286/2016/$25.00 © 2016 SPIE

Optical Engineering 013101-1 January 2017 • Vol. 56(1)

Optical Engineering 56(1), 013101 (January 2017)

data were obtained from Riegl VZ-400. The detailed infor-mation about ALS and TLS measurements and instrumentsare given in Table 1.

3 MethodologyFor the validation of the ALS accuracy, five main steps wereused (Fig. 2). The common commercial software package,

TerraSolid for MicroStation (v14), was chosen to process theALS data in .las format.19,20 For the generation of DSMs, 3-Dtransformation, and accuracy analysis, Surfer (v11) and theprogram system Bundle Block Adjustment Leibniz Univer-sity of Hannover (BLUH) were utilized. In the course of theanalysis, the coordinate system was UTM zone 35 based onWGS84 datum.

3.1 Improvement of Airborne Laser Scanning PointClouds

The acquisition of ALS point clouds is based on direct sensororientation and boresight alignment. For the relative kin-ematic global navigation satellite system (GNSS) positioningof direct sensor orientation, at least eight satellites with aposition dilution of precision value below 2.5 were usedfor every flight line, and the processing indicated that thestandard deviation (σz) in the Z-direction was between 4and 7 cm. In addition, the boresight alignment, laser scanner,and definition of an object “point” had an impact on the data.Crossing flight lines can be used for the block handling ofALS strips supported by ground control points (GCPs), butin this project, such crossing flight lines were not availableand for the high number of linear flight lines, collecting aGCP for every line is not realistic. Therefore, subblocks of∼10 flight lines were built. With TerraMatch included in theTerraSolid application, ∼50% of the lateral overlappingstrips of the different flight lines were three-dimensionallytransformed together and the areas were monitored basedon the extracted inclined planes with at least a 10-deg

Fig. 1 Location of the test areas in Istanbul.

Table 1 Specifications of the ALS and TLS measurements and instruments.

Parameter Airborne laser scanning TLS areas 4 to 7

Instrument Riegl LS Q680i Riegl VZ-400

Dates of flight/ground survey Area 4: October 12, 2012 Area 4: July 16 to 17, 2012

Area 5: September 30, 2012 Area 5: August 17, 2013

Area 6: September 23, 2012 Area 6: July 10, 2013

Area 7: October 3, 2012 Area 7: June 30 + July 10, 2013

Flying altitude 600 m —

Laser footprint 30 cm at 600 m 3 cm at 100 m

Average point density (per 1 m2) 20 ∼10;000 points ≤ 10 m

Field of view (deg) −30 deg to þ30 deg 360 deg horizontal∕ − 40 deg to þ60 deg vertical

Pulse rate 400 kHz 300 kHz

Wavelength 1550 nm 1550 nm

Scan frequency 200 Hz N/A

Beam divergence 0.5 mrad 0.35 mrad

Number of returns 4 4

Velocity of helicopter 150 km∕h —

Integrated mid-format digital camera IGI DigiCam 60 (50-mm focal length) —

Optical Engineering 013101-2 January 2017 • Vol. 56(1)

Sefercik et al.: Point-based and model-based geolocation analysis of airborne laser scanning data

inclination. Such inclined planes, usually roofs of buildings,allow not only vertical but also horizontal fitting.21 Suchplanes are not much affected by roughness.

3.2 Point-Based Geolocation Accuracy Analysis

Point-based accuracy analysis is the simplest way to deter-mine the vertical system accuracy of ALS point clouds.22

However, considering the nearly continuous nature of anALS point cloud, it is not possible to estimate the absoluteaccuracy of point clouds in a rough terrain using a limitednumber of checkpoints. On this basis, point-based geometrictest of the merged ALS point clouds was utilized to estimateALS geolocation accuracy in flat areas. The process wasconducted with 17 checkpoints on a tennis court (Fig. 3).

The expected accuracy depends upon the σz of the laserrange of nominal {plus minus} 2 cm, the accuracy of the

scan angle {plus minus} 0.003 deg, the inertial measurementunit (IMU) roll and pitch {plus minus} 0.003 deg, the IMUheading {plus minus} 0.01 deg, and the relative kinematicpositioning {plus minus} 5 cm. For the {plus minus}30-deg field of view, the influence of the relative kinematicpositioning on the height of the determined object was in thesame range as all the effects of the attitudes at the ALS striplimit for the 600-m flying elevation, leading to an approxi-mate σz of 7.5 cm in the Z-direction. This was improvedby fitting and merging the overlap areas of the neighboringsubblocks of strips by GCPs.

3.3 Model-Based Geolocation Accuracy Analysis

Model-based analysis is based on the differences between theALS DSMs and the reference DSMs, rather than the use ofa limited number of checkpoints. This is a more realistic

Fig. 2 Methodology.

aeratseTRMSEsnoitairaV

ALS DSM – checkpoints 3.3 cm (bias=0.3 cm)

height variation of thecheckpoints determined by GNSS

+/- 3 cm

height variation of the ALS data

+/- 1.4 cm

Fig. 3 Checkpoint distribution on the tennis court.

Optical Engineering 013101-3 January 2017 • Vol. 56(1)

Sefercik et al.: Point-based and model-based geolocation analysis of airborne laser scanning data

technique for the accuracy analysis of a DSM.23 Thereference model has to be achieved using the same or a moreprecise technique as that of the tested model, and its originalgrid spacing has to be at least equal or smaller. Consideringthese criteria, ALS and TLS data are the most appropriatereferences for the analysis of the ALS data.24 In this study,DSMs, derived from ∼50% overlapping ALS strips, werecompared internally. In addition, DSMs, achieved frommatched and merged ALS strips, were checked against theTLS DSMs.

In addition to σz, the normalized median absolutedeviation (NMAD) [Eq. (2)] was used as an accuracy cri-terion. NMAD is the derivative of median absolute deviation(MAD) [Eqs. (1) and (2)]. σz and NMAD are the main accu-racy criteria facilitating the validation and specification ofoutliers. Both accuracy figures are identical if the evaluateddiscrepancies are normally distributed; however, this is usu-ally not the case.25 In Eq. (1), x̃i is the median of absoluteheight discrepancies between the two DEMs.

σz is related to average while root-mean-square errors(RMSE) are influenced by the bias. Due to the square sumnature of σz, it is strongly affected by larger discrepancies.NMAD is based on the median, so it does not change muchif the error frequency distribution does not correspond toa normal distribution. However, the normal distribution ofthe discrepancies is usually limited, in which case NMADdescribes the frequency distribution of discrepancies better,usually being smaller than σz. The correct description of thevertical accuracy of a height model is based on the Euclidiandistances, the shortest distance between height models,depending upon the cosine of the terrain inclination. In ourinvestigation, the Euclidian standard deviation was nearlythe same as σz in the vertical direction. More than thoseused in this study, several estimators such as relative bias,average relative absolute difference (ARAD), and logRMSE are generally utilized for the accuracy validation ofDEMs derived from remotely sensed data.26

EQ-TARGET;temp:intralink-;e001;63;345MAD ¼ x̃i½jΔZij�; (1)

EQ-TARGET;temp:intralink-;e002;63;313NMAD ¼ 1.4815 ×MAD: (2)

At the first step of the model-based analysis, in threestudy areas, 9 DSMs were generated in nine strips andvalidated by being comparing with each other in ∼50% over-lapping parts of the strips. In the second step, completingthe TLS measurements in four areas, the reference data wereobtained for the merged ALS data. Figure 4 shows the ALSclassification (color-coded) and TLS point-clouds (black andwhite) in the four test areas with orthometric height scales. Inarea 4, the TLS data were obtained from mobile mapping.The ALS and TLS DSMs were generated by TerraScan inTerraSolid. Required interpolations were based on kriging.27

The horizontal offsets between the ALS and TLS DSMswere eliminated by shifting based on an area-matching crosscorrelation.

In addition to the absolute standard deviation (Aσz),relative standard deviation (Rσz) indicating the interiorconsistency of a DSM based on the relationship betweenthe neighboring grids was calculated separately for eacharea. The distance of the current point to a reference pointindicates height differences, which are compared for all

neighboring grids. In the case of independent heightdifferences (correlation ¼ 0), Rσz would be identical to σz,and if the correlation coefficient is 1.0, Rσz would be zero.Therefore, under normal conditions, Rσz is between zero andσz. However, in the extreme case of a negative correlation,the Rσz may be larger than σz. Using Rσz, the dependency(correlation) of the height differences between the tested andreference models (DZ) is checked, through which localsystematic errors and random errors can be differentiated.It has been reported that Rσz of active remote sensing sys-tems such as radar and lidar are superior to their Aσz.28–30

Rσz of the DSMs were calculated using Eq. (3), where Drepresents the distance groups and Dl and Du are the lowerand upper distance limits. In this study, for each pixel, theneighboring 10 pixels were used in the calculation; therefore,the distance groups varied from the first to the tenth pixel(1 to 10 m based on a 1-m pixel size of the referencedata and 0.5 to 5 m for 0.5 m reference data). nx indicatesthe number of point combinations in the distance group andn represents the total number of differences between thereference and actual values. A factor of 2 for multiplicationwith nx is used for norming Rσz to the size of σz. If the

Fig. 4 Point clouds in the test areas 4 to 7; ALS terrain classificationusing absolute elevation (in color), TLS point clouds (black and white).

Optical Engineering 013101-4 January 2017 • Vol. 56(1)

Sefercik et al.: Point-based and model-based geolocation analysis of airborne laser scanning data

height differences of the points in the distance group areindependent, corresponding to the error propagation, thenthe RMS would be the square root of 2.0 larger than σz,which is represented by 2 × nx

EQ-TARGET;temp:intralink-;e003;63;117Rσz ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiP ðDZi −DZjÞ2

ð2 × nxÞ

sand Dl < D < Du: (3)

As expected, height models and the laser scan data are not asaccurate in an inclined terrain as in flat areas. As known,when generating any kind of DEM, the interpolation isa crucial step for vector-raster transformation to assigna gray value for each pixel of the raster model. On theother hand, the interpolation causes geolocation accuracyloss on the raster model due to produce derivative gray val-ues, which are not as reliable as the original point-clouds.The influence of interpolation in final 3-D product depends

Fig. 5 0.5-m gridded DSMs of overlapping ALS strips for the test areas 1 to 3 (strips 1, 2, 3 = area 1;strips 4, 5, 6 = area 2; strips 7, 8, 9 = area 3).

Optical Engineering 013101-5 January 2017 • Vol. 56(1)

Sefercik et al.: Point-based and model-based geolocation analysis of airborne laser scanning data

on several parameters such as original grid spacing of thedata, interpolation window size, used interpolation method,the characteristics of land cover (open, forest, etc.) and theinclination of the terrain.31 In this study, the effect of slope onthe accuracy was taken into account by σz ¼ aþ b × tan(terrain inclination) with a and b indicating the certain accu-racy value and the factor of terrain inclination, respectively.The analysis describes the accuracy corresponding to thisdependency. In this analysis, only height discrepancieswithin a realistic threshold are taken into account, theremaining data was handled as blunders. Here, a threshold of{plus minus} 1 m between ALS DSMs and reference DSMswas used and the percentage of blunders in all analyzed datawas included in accuracy tables as excluded points. Blundersmay be caused by growing vegetation related to the temporalspacing between the ALS and TLS data collection dates, andthe different viewing geometry of the ALS and TLS scansshowing different object parts in vegetation.

4 Results

4.1 Results of Point-Based Analysis

Figure 3 shows the RMS differences of ALS points against17 checkpoints on the tennis court and the height variationsof the checkpoints.

The small bias is a random result. However, the RMSdifferences in height (RMSE: 3.3 cm) and the internalheight variation of ALS heights on the flat tennis court beingonly 1.4 cm indicate high relative accuracy. The height varia-tion of the GNSS heights being 3 cm indicates lower accu-racy of the reference measurement compared to the ALSheights.

4.2 Results of Model-Based Analysis

Figure 5 shows the overlapping ALS DSMs of the test areas1, 2, and 3 with 50 cm grid spacing (color-coded). The directcomparison of the overlapping ALS heights is dominated bythe random influence of vegetation and building outlines. Atbuilding outlines, points with the same X- and Y-locationmay be located on the roof overhang in one strip andon the facade in the other. Table 2 shows the results of

∼50% overlapped ALS strips after shift correction withRiPROCESS in TerraSolid. Strips 1 to 3 are not the same asthe test areas used for the comparison with the TLS data. Thecomparison of strip 7 with strip 8 indicates a clear rotation ofthese strips against each other. After the correction of thisrotation, σz and NMAD were within the usual range. Thevalues in Table 2 are clearly greater than the results obtainedfrom the tennis court due to object roughness. Object rough-ness is not a problem for strip transformation by TerraSolidsince it is based on inclined flat planes, e.g., building roofs.

Figure 6 shows the ALS and TLS DSMs in areas 4 to 7after shift correction based on area-matching cross-correla-tion. The horizontal shifting values were found to rangefrom −3.08 to 1.07 cm in the X-direction and from −6.00 to0.18 cm in the Y-direction. The grid spacing of the DSMswas 1 m with the exception of area 6 where it was 0.5 m.

Figure 7 shows, in white, the points (pixels) that wereexcluded due to exceeding the 1-m threshold between theALS and TLS. Particularly in vegetation, due to unstablegeometry, different footprint locations and view directionslead irregular height values in point definition. Similar prob-lems also exist on facades and roof overhang points. Movingcars are another reason for the excluding threshold. Apartfrom the exclusion method by determining a threshold,problematic points can be removed by using an appropriatefiltering algorithm.

The frequency distributions of height differences betweenthe ALS DSMs and TLS DSMs in Fig. 8 show the nonne-gligible bias, requiring an optimal fit of the TLS to ALS data.In addition, the limited normal distribution is demonstrated.The large differences between σz and NMAD are caused bythe higher number of larger discrepancies, influencing thesquare nature of σz more than the normalized median.Figure 9 shows the Rσz of height differences in areas 4,5, 6, and 7. The results vary from 12 to 25 cm, which indi-cates normal conditions for all the study areas. The correla-tion of neighboring pixels is very local with Rσz notchanging in distances exceeding 5 or less pixels. The resultsare superior to Aσz as expected for active remote sensingsystems.

With respect to the guidelines published by the AmericanSociety for Photogrammetry and Remote Sensing (ASPRS),the horizontal and vertical accuracy of the ALS DEMs wereevaluated based on the produced map scale and contourinterval according to the mapping requirements determinedby United States National Map Accuracy Standards (NMAS)and National Standards for Spatial Data Accuracy (NSSDA)(Table 4). Horizontal shifting and model-based absolutevertical accuracy after the elimination of systematic biasdemonstrate that the ALS DSMs fulfill the horizontal andvertical accuracy criteria of 1∕1000 scaled topographic mapsat the 95% confidence level (compare Tables 3 and 4).

5 ConclusionIn this study, the geo-location characteristics and accuracy ofthe ALS DSMs were analyzed using point-based and model-based approaches. The point-based analysis was conductedon a completely flat tennis court using checkpoints. Thisallowed the verification of the absolute and relative heightlevel of the ALS height model but not the confirmation of thehorizontal location. Absolute and Rσz of the height were notrepresentative for an undulating or rough terrain. Through

Table 2 Accuracies of ALS DSMs derived from the combinations ofALS strips.

Strip combinations σz (cm) NMAD (cm)

Influence of roll toZ from one side to

another (cm)

1 to 2 6.0 5.0 2.5

2 to 3 10.0 9.0 2.8

4 to 5 5.0 4.0 0.0

5 to 6 6.0 4.0 2.9

7 to 8 12.0 10.0 21.0

8 to 9 8.0 6.0 5.0

7 to 8 afterrotation correction

8.0 7.0 1.5

Optical Engineering 013101-6 January 2017 • Vol. 56(1)

Sefercik et al.: Point-based and model-based geolocation analysis of airborne laser scanning data

the model-based analysis of the overlapping ALS strips,more realistic accuracy values were obtained, but the differ-ent viewing directions required the exclusion of vegetationareas as well as building outlines. The model-based analysisbased on the comparison of ALS DSMs of the merged stripswith the TLS DSMs presented similar problems in terms ofthe vegetation and building outlines, which required filtering.

The difference between σz of height and NMAD demon-strated a limited normal distribution of the height discrepan-cies, influencing σz more than NMAD. In addition, NMADwas found to demonstrate the frequency distribution of theheight differences better than σz, due to the latter stronglydepending upon larger discrepancies. It was clear that theaccuracy depended upon terrain inclination, which required

Fig. 6 Color-coded DSMs (left) and differential DEMs with bias (right); height and height differencescale (m).

Optical Engineering 013101-7 January 2017 • Vol. 56(1)

Sefercik et al.: Point-based and model-based geolocation analysis of airborne laser scanning data

the description of accuracy as a function of terrain inclina-tion. In general, the vertical accuracy of the ALS DSM wasaround 3 cm in flat areas such as the tennis court, and NMADranged from 7 to 14 cm for a vegetated and rough terrain.

The model-based analysis also facilitated the determina-tion of the horizontal position of the height model, whichcannot be determined in flat areas. Overall, the large ALSDSM based on helicopter flights without crossing flightlines lead to satisfying results, confirming the strategy used.In areas of vegetation, the discrepancies between differentdatasets are dominated by the point definition caused by thefootprint location not being the same and different viewdirections leading to varying height values also being pointson facades and roof overhangs. In a rough terrain, such ascultivated land with low vegetation and uncultivated areas,the RMS differences between the height determination from

Fig. 7 Excluded points (in white) where ΔZ > 1 m.

Fig. 8 Frequency distribution of height differences between the ALS and TLS DSMs before and after theoptimal fitting of strips.

Fig. 9 Rσz of the ALS DSMs.

Optical Engineering 013101-8 January 2017 • Vol. 56(1)

Sefercik et al.: Point-based and model-based geolocation analysis of airborne laser scanning data

different flight lines and between ALS and TLS data moreclearly show the definition of the object surface as the accu-racy of laser scanning. The comparison of the ALS and TLSresults is an excellent way of obtaining realistic informationabout object definition, especially in vegetated and roughareas as well as at building outlines. In terms of the ASPRSguidelines, the ALS DSMs fulfill the horizontal and verticalaccuracy requirements of 1∕1000 scaled topographic maps atthe 95% confidence level.

References

1. E. Baltsavias, “A comparison between photogrammetry and laserscanning,” ISPRS J. Photogramm. 54(2), 83–94 (1999).

2. J. Hill et al., “Wide-area topographic mapping and applications usingairborne light detection and ranging (Lidar) technology,” Photogramm.Eng. Remote Sens. 66(8), 908–914 (2000).

3. U. Lohr, “Digital elevation models by laser scanning,” Photogramm.Rec. 16(91), 105–109 (1998).

4. J. Shan and A. Sampath, “Urban DEM generation from raw lidar data:a labelling algorithm and its performance,” Photogramm. Eng. RemoteSens. 71(2), 217–226 (2005).

5. X. Liu, “Airborne LiDAR for DEM generation: some critical issues,”Prog. Phys. Geog. 32, 31–49 (2008).

6. M. Hollaus et al., “Airborne laser scanning and usefulness for hydro-logical models,” Adv. Geosci. 5, 57–63 (2005).

7. M. D. McCoy et al., “Airborne lidar survey of irrigated agricultural land-scapes: an application of the slope contrast method,” J. Archaeol. Sci.38(9), 2141–2154 (2011).

8. L. Jing et al., “Automated delineation of individual tree crowns fromlidar data by multi-scale analysis and segmentation,” Photogramm.Eng. Remote Sens. 78(12), 1275–1284 (2012).

9. F. M. Southee et al., “Application of lidar terrain surfaces for soil mois-ture modeling,” Photogramm. Eng. Remote Sens. 78(12), 1241–1251.

10. K. D. Fieber et al., “Validation of canopy height profile methodology forsmall-footprint full-waveform airborne LiDAR data in a discontinuouscanopy environment,” ISPRS J. Photogramm. Remote Sens. 104, 144–157 (2015).

11. A. Khosravipour et al., “Generating pit-free canopy height models fromairborne lidar,” Photogramm. Eng. Remote Sens. 80(9), 863–872 (2014).

12. L. Cao et al., “Tree species classification in subtropical forests usingsmall-footprint full-waveform LiDAR data,” Int. J. Appl. Earth Obs.Geoinf. 49, 39–51 (2016).

13. N. Csanyi and C. K. Toth, “Improvement of lidar data accuracy usinglidar-specific ground targets,” Photogramm. Eng. Remote Sens. 73(4),385–396 (2007).

14. S. Dahlqvist et al., “Evaluating the correctness of airborne laser scan-ning data heights using vehicle-based RTK and VRS GPS observa-tions,” Remote Sens. 3(12), 1902–1913 (2011).

15. C. Hladik and M. Alber, “Accuracy assessment and correction ofa LIDAR-derived salt marsh digital elevation model,” Remote Sens.Environ. 121, 224–235 (2012).

16. W. T. Tinkham et al., “Amethodology to characterize vertical accuraciesin Lidar-derived products at landscape scales,” Photogramm. Eng.Remote Sens. 79(8), 709–716 (2013).

17. R. Canavosio-Zuzelski et al., “Assessing Lidar accuracy with hexagonalretro-reflective targets,” Photogramm. Eng. Remote Sens. 79(8), 663–670 (2013).

18. S. Gould et al., “Influence of a dense, low-height shrub species on theaccuracy of a LiDAR-derived DEM,” Photogramm. Eng. Remote Sens.79(5), 421–431 (2013).

19. M. Isenburg, “LASzip,” Photogramm. Eng. Remote Sens. 79(2), 209–217 (2013).

20. A. Khosravipour et al., “Effect of slope on treetop detection usinga LiDAR canopy height model,” ISPRS J. Photogramm. Remote Sens.104, 44–52 (2015).

21. J. Rentsch and P. Krzystek, “Lidar strip adjustment with automaticallyreconstructed roof shapes,” Photogramm. Rec. 27(139), 272–292 (2013).

22. J. Höhle, “Assessing the positional accuracy of airborne laser scanningin urban areas,” Photogram. Rec. 28(142), 196–210 (2013).

23. Q. Lin et al., “Comparison of elevation derived from InSAR data withDEM over large relief terrain,” Int. J. Remote Sens. 15, 1775–1790(1994).

24. A. P. Young et al., “Comparison of airborne and terrestrial lidarestimates of seacliff erosion in southern California,” Photogramm. Eng.Remote Sens. 76(4), 421–427 (2010).

25. J. Höhle and M. Höhle, “Accuracy assessment of digital elevationmodels by means of robust statistical methods,” ISPRS J. Photogramm.Remote Sens. 64, 398–406 (2009).

Table 4 Map accuracy requirements based on the map scale and contour interval.29

Map scale Contour interval (m)

Horizontal Vertical

RMS (cm)Accuracy (95% confidence level)

(1.96 × RMS) (cm) RMS (cm)Accuracy (95% confidence level)

(1.96 × RMS) (cm)

1∶500 0.5 0.28 0.55 4.60 9.10

1∶1000 1 0.56 1.10 9.25 18.2

1∶2000 2 1.12 2.20 18.5 36.3

1∶5000 5 2.79 5.47 46.3 90.8

Table 3 Absolute vertical accuracy of ALS DSMs in relation to TLS DSMs (α ¼ slope).

Area Spacing (m) Number of points Bias (m) RMSE (m) σz (m) NMAD (m) σz as a function of slope (m) Excluded points (%)

4 1 59,839 0.15 0.23 0.18 0.07 0.22þ 0.10 × tanðαÞ 2.84

0.00 0.19 0.19 0.07 0.21þ 0.07 × tanðαÞ 2.36

5 1 4569 0.00 0.25 0.25 0.10 0.23þ 0.09 × tanðαÞ 15.77

6 0.5 1774 0.04 0.25 0.25 0.03 0.10þ 0.30 × tanðαÞ 21.33

0.00 0.25 0.25 0.03 0.08þ 0.35 × tanðαÞ 21.24

7 1 5336 0.05 0.19 0.18 0.08 0.14þ 0.34 × tanðαÞ 20.63

0.00 0.19 0.19 0.08 0.15þ 0.26 × tanðαÞ 20.52

Optical Engineering 013101-9 January 2017 • Vol. 56(1)

Sefercik et al.: Point-based and model-based geolocation analysis of airborne laser scanning data

26. S. P. Wechsler and C. N. Kroll, “Quantifying DEM uncertainty and itseffect on topographic parameters,” Photogramm. Eng. Remote Sens.72(9), 1081–1090 (2006).

27. C. S. Yang et al., “Twelve different interpolation methods: a case studyof surfer 8.0,” in Proc. of the ISPRS XXth Congress, Istanbul, Turkey,pp. 778–783 (2004).

28. K. Jacobsen, “DEM generation from satellite data,” in EARSeL (Hrsg.),p. 4, Millpress, Ghent (2003).

29. ASPRS (American Society of Photogrammetry and Remote Sensing),ASPRS Guidelines Vertical Accuracy Reporting for Lidar Data,2014, http://www.asprs.org/a/society/committees/lidar/Downloads/Vertical_Accuracy_Reporting_for_Lidar_Data.pdf (15 November 2014).

30. K. Jacobsen, “Characteristics of airborne and space-borne heightmodels,” in Proc. of the EARSeL Symp., Matera, Italy (2013).

31. A. L. Montealegre et al., “Interpolation routines assessment in ALS-derived digital elevation models for forestry applications,” RemoteSens. 7, 8631–8654 (2015).

Umut Gunes Sefercik received his PhD fromBulent Ecevit University(BEU), Turkey, in 2010. He finished his postdoctoral at the NationalCenter for Airborne Laser Mapping (NCALM), University of Houston,USA, in 2014. He is an associate professor at the Department of

Geomatics Engineering, BEU. His research interests are processing,generation, and quality validation of airborne and spaceborne remotesensing data and products.

Gurcan Buyuksalih received his PhD from the Department ofGeography and Topographic Science, University of Glasgow, UK.He is an associate professor in photogrammetry at Bimtas, Istanbul,Turkey. His research direction is the full range of photogrammetry,especially application of space imagery.

Karsten Jacobsen received his PhD in photogrammetry from LeibnizUniversity Hannover, Germany. His main research area is numericalphotogrammetry, especially the use of space imagery.

Mehmet Alkan received his PhD fromKaradeniz Technical University(KTU), Turkey, in 2005. He is an associate professor at the Depart-ment of Geomatics Engineering, Yıldız Technical University. Hisresearch interests are database management, geographical informa-tion systems, remote sensing, object extraction, national spatial datainfrastructure, e-municipality, and e-government.

Optical Engineering 013101-10 January 2017 • Vol. 56(1)

Sefercik et al.: Point-based and model-based geolocation analysis of airborne laser scanning data